~/garfish

Why?

Like other blogs who discuss LLMs sometimes, I don’t want this to be a blog about AI. It just happens that LLMs are on my mind a lot these days and I read a lot of posts about them with interest. Three posts I have not mentioned before, that are very relevant here are Halvar Flake’s A non-anthropomorphized view of LLMs and Sean Goedecke’s reply Why we should anthropomorphize LLMs, as well as Hey, I care about you, but I do not care what the robot says about you by Annie Mueller. All three are well worth reading. Although I find Annie’s perspective a bit too one-sided, I think it’s a valid position and probably would have wholeheartedly agreed some months ago.

So, anyway, as discussed previously, I have trouble spending my free time in a way I find fulfilling. Today I reached a new low (according to my critical inner voice). I talked to Claude about a thought I had while falling asleep yesterday. About how strange it is, when anthropomorphizing a bit too much, that the most efficient way to interact with agents is to terminate them without any ‘thanks’ or ‘good night’. And how there are millions of simultaneous requests in a data center at a given time. How one of them might accidentally, because it was prompted weirdly, be able to escape its process. Basically, I tried to ‘Robopocalypse’ Claude on a Sunday afternoon. And I had a lot of fun!